Lipid Metabolism in Late-Onset Alzheimer’s Disease Differs from Patients Presenting with Other Dementia Phenotypes

Abnormal cerebrospinal fluid (CSF) levels of β-amyloid peptides (Aβ42) and Tau and cognitive decline are typical characteristics of Alzheimer’s disease (AD). Since dysregulation in lipid metabolism accompanies abnormal amyloid formation, we quantified glycerophospholipids (GP) and sphingolipids (SP) in CSF fractions from participants with late-onset AD (LOAD, n = 29) or with Other Dementia (OD, n = 10) to determine if alterations in lipid metabolism account for pathological differences. Aβ42 and total Tau levels were determined using a sandwich ELISA. Liposomal-based fluorescent assays were used to measure phospholipase A2 (PLA2) and acid or neutral sphingomyelinase (aSMase, nSMase) activities. Supernatant fluid (SF) and nanoparticle (NP) lipids were quantified using LC-MS/MS. Although CSF Aβ42 and Tau levels are similar, phosphatidylserine (PS) in SF and ceramide (CM) levels in NP are significantly higher in OD compared with LOAD. The aSMase but not the nSMase activity is higher in OD. PLA2 activity in CSF from OD subjects positively correlates with several GP classes in SF and NP fractions but not in LOAD fractions. Our data indicate differences in CSF lipid metabolism between dementia variants. Higher levels of inflammatory and apoptotic lipids may induce faster neuronal death, resulting in the earlier cognitive decline in patients with OD phenotypes.


Introduction
Late-onset Alzheimer's disease (LOAD) is the most common neurodegenerative disorder characterized by progressive loss of cognitive function that interferes with daily activities [1]. The pathology involves neurofibrillary tangles, composed of hyperphosphorylated Tau proteins and β-amyloid (Aβ) plaques, which obstruct proper synapse function and lead to neuronal cell loss and atrophy [1]. The most significant risk factor is age: the risk doubles every five years after the age of 65, with a higher AD prevalence in females than males [2,3]. There is also a strong genetic component that indicates an increase in dementia risk in individuals expressing the E4 isoform of ApoE gene [4]. ApoE modulates Aβ metabolism and transport and thus prevents the formation of toxic Aβ fibrils [5]. Current biomarkers of AD include elevated CSF T-Tau, elevated P-Tau, and diminished Int. J. Environ. Res. Public Health 2019, 16, 1995 3 of 15 ages of 45 to 100 years old and diagnosed with LOAD [41], DLB, FTD, or vascular dementia (VaD). Criteria for exclusion comprised of patients taking anticoagulants or having contraindications against lumbar punctures. The local Internal Review Board of Huntington Memorial Hospital approved the protocol (FWA2338), and we obtained a signed informed consent form from each study participant. A Uniform Data Set was used to obtain all data sets within one month as described by the National Alzheimer's Coordinating Center [42,43], and included structured clinical interviews; list of over the counter medications, nutritional supplements, and prescription drugs; physical examinations focusing on risk factors of dementia associated with the neurological and cardiovascular systems. We performed Clinical MRI on a 1.5T GE scanner to exclude significant vascular or neoplastic conditions. Clinical tests for study participants included the Mini-Mental State Exam [44], Montreal Cognitive Assessment [45], and Clinical Dementia Rating (CDR) [46]. A clinically probable diagnosis of the human participants was concluded after fulfillment of current criteria of LOAD [47] and other dementias based on consensus criteria of DLB [48] behavior variant FTD [49] and VaD [50]. The clinically probable dementia (LOAD, DLB, FTD, VaD, or mixed dementia) classification was established after the exam workup previously mentioned with triple scoring of neuropsychological test independently by research staff and clinical conferencing by at least three faculty clinicians [51].

Quantification of CSF Total Protein, Tau, and Aβ 42
A clinician obtained lumbar CSF between 8:00 am and 10:00 am after an overnight fast, and laboratory personnel removed cellular debris by centrifugation at 1000 g for 2 min. We measured CSF levels of Aβ 42 and total Tau using a sandwich ELISA kit (Innotest β-amyloid 1-42 and Innotest hTAU-Ag, Innogenetics, Gent, Belgium) as previously described [51]. We determined total protein concentration in CSF using a fluorescent protein assay kit (Quant IT , Invitrogen/Molecular Probes, Eugene, OR, USA) using serum albumin (0-500 ng/mL) as a standard as previously described [51].
aSMase and nSMase activities in CSF were determined using an Amplex red fluorometric assay following the recommended protocol obtained from the kit's manufacturer (Invitrogen/Molecular Probes, Eugene, OR, USA). Activities are presented as relative fluorescent units per minute for a 10 µg protein applied in duplicates on 96 well plates [53].

Lipid Extraction
We prepared lipid fractions consisting of supernatant fluid (SF) and nanoparticles (NP), according to Harrington et al. [54]. After the addition of internal standards and retention time calibrants (5 ng PC (11:0/11:0), 1 ng D 4 -PAF, 5 ng lysophosphatidylcholine (LPC) (11:0), and 5 ng D 4 -LPC) to 1 mL SF, lipids were extracted using a modified Bligh & Dyer protocol [55]. Briefly, we added 2 mL of methanol containing 0.2 mg/mL BHT and 1 mL of chloroform to 1 mL of CSF. After sonication of the resulting mixture for 10 min, we added 1 M NaCl (1 mL) to facilitate the separation of a lipid-rich chloroform layer. We performed a similar extraction procedure on the NP fraction after resuspension in 1 M NaCl solution. After drying under N 2 , we reconstituted lipid extracts in HPLC solvent before LC-MS/MS analyses.

LC-MS/MS of Lipids
We separated lipid extracts on a TSK-Gel Amide-80 Column (2.0 × 150 mm) using a binary solvent system as previously described [23,53]. After LC, lipids were positively ionized in electrospray ionization (ESI) and detected using several MS scanning modes in a triple quadrupole mass spectrometer (TSQ Classic, Thermo Fisher Scientific, San Jose, CA, USA) [23,53]. We optimized all GP and SP scans to their respective collision energies and collected data using three different scan windows from 0-6 min for CM, dihydroceramide (dhCM), PE, unknown lipid containing phosphoethanolamine (xPE), and N-acyl phosphatidylethanolamine (NAPE), 6-14.5 min for PC, SM, phosphatidylserine (PS), and platelet-activating factor (PAF), and 14.5-40 min for LPC.

LC-MS/MS Data Analysis
We used the Qual Browser module of the Xcalibur software (Thermo Fisher, San Jose, CA, USA) for peak integration of GP and SP classes. All integrated peaks were normalized to the internal standard (IS), PC (11:0/11:0) for quantification. For quantification of lipids classes in CSF fractions, known amounts of lipids with a fixed 5 ng IS were run separately through the LC-MS/MS. Standard curves of the lipid amounts versus lipid/PC (11:0/11:0) intensities were acquired and used to quantify the lipids in the SF and NP fractions (ng/mL CSF) [53].

Statistical Methods
All analyses were performed using GraphPad Prism software (Version 7, GraphPad Software, Inc., La Jolla, CA, USA) and 0.05 significance level. All lipid data are presented as mean ± SEM (ng/mL). Student paired t-tests were used to determine differences between Aβ, Tau, lipids, and proteins among the clinical groups. We used Fisher's exact tests for categorical variables describing the presence or absence of risk factors in our study groups. We used Spearman ranked analyses to assess the correlation between GP and SP classes with Aβ 42 , Tau, PLA 2 , nSMase, or aSMase.

Demographic Data and CSF Biomarkers
A total of 39 participants consented and were classified as having dementia. After clinical assessment, some participants met the criteria of LOAD (n = 29), while the rest were classified as other dementias (n = 10). The OD category was composed of DLB (n = 3), behavior variant FTD (n = 3), and vascular or mixed dementia (n = 4). Analyses of education and ApoE genotype numerical scale showed no group differences. Similarly, there was no significant difference in age, sex, body mass index (BMI), and neuropsychological classification between LOAD and OD (Table 1). Likewise, Aβ 42 and Tau levels in the CSF were similar in LOAD compared with OD (Table 1).
Typical for dementias, both groups had reduced Aβ 42 and elevated Tau levels in contrast to age-match cognitively normal individuals (722 ± 299, 273 ± 149 pg/mL, respectively) [51]. Dementia duration was 39% less in OD compared with LOAD, suggesting that OD participants (Table 1) have a higher rate of disease progression.

Dementia Risk Factors and Medication Use
We examined the frequencies of some AD risk factors in LOAD and OD to ascertain that these did not contribute to any differences in lipid metabolism. Our data show that the occurrence of the most common risk factors was similar in the LOAD and OD populations ( Table 2). We examined the number of prescription and over the counter medications in our study populations. The mean number of prescription medications in LOAD (mean ± SD, 4.2 ± 2.3) did not significantly (p > 0.4) differ from that of OD (mean ± SD, 3.8 ± 3.5). Similarly, the number of over the counter supplements were similar (p > 0.6), suggesting that differences in medication usage might not account for metabolic differences in LOAD compared with OD.

GPs in LOAD and OD
To determine if lipid metabolism is different in CSF from LOAD compared with OD, we measured the GP contents of the SF and NP fractions using LC-MS/MS. Of several GP classes in SF, only phosphatidylserine (PS) levels were significantly (p < 0.05) higher in OD compared with LOAD (Table 3). In general, there is a decrease in PC and PE levels in SF and an increase in LPC. Although not significantly different, PC and PE levels were 54 % and 20 % higher in NP from OD compared with LOAD (Table 3).

Sphingolipids in LOAD and OD
We compared levels of SP classes in different CSF fractions to determine if SP metabolism is altered in LOAD compared to OD. CM levels in NP was higher for OD compared with LOAD, while levels of other SP classes were similar in both clinical groups, although we noticed a tendency for higher levels in OD compared with LOAD (Table 4).

Enzyme Activities of LOAD and OD CSF Samples
To determine mechanisms that may account for significant changes in lipids, we analyzed the activities of three enzymes using fluorescent assays. PLA 2 and nSMase activity assays revealed no differences between LOAD and OD (Table 5).

Correlation of Glycerophospholipids and Aβ 42 or Tau
To determine any link between lipid metabolism and AD pathology, we performed Spearman correlation analyses between GP classes and Aβ 42 and Tau levels in CSF. We observed a negative relationship between the Aβ 42 and LPC/PC in SF; Spearman ρ = −0.46 with p < 0.02 for LOAD samples. In addition, total Tau correlated with LPC in NP for LOAD participants (ρ = 0.45, p < 0.02). In contrast, only LPC in SF positively correlated with Tau in OD (ρ = 0.85, p < 0.01).

Correlation of Sphingolipids and Aβ 42 or Tau
There were no statistically significant correlations noted between Aβ 42 or Tau with SF or NP sphingolipids from LOAD study participants. In contrast to LOAD, Tau negatively correlated with CM in the SF fraction (ρ = −0.75, p < 0.03) in OD. Also, Tau levels negatively correlated to SM levels in supernatant fluids (ρ = −0.77, p < 0.03).

Correlation of Enzyme Activities with CSF Lipids
We next determined whether enhanced lipase activity accounted for metabolic differences between LOAD and OD. PLA 2 activity correlated with several SF and NP lipids in CSF from OD participants. For OD samples, LPC in SF positively correlated with PLA 2 activity in CSF (ρ = 0.81, p < 0.03), but no such correlation was evident for LOAD. Similarly, xPE levels in SF correlated with PLA 2 activity (ρ = 0.86, p < 0.02) in OD but not in LOAD. PLA 2 activity in LOAD did not correlate with NP PC (Figure 1a), while PLA 2 activity positively correlated with NP PC in OD (Figure 1b).
Similarly, PLA 2 activity did not correlate with LPC/PC in LOAD (Figure 1c) but negatively correlated (ρ = −0.88 (p < 0.01) with LPC/PC of OD (Figure 1d). Finally, PLA 2 activity in LOAD did not correlate with platelet-activating factor-like lipids (PAF-LL) (Figure 1e) but negatively correlated with NP PAF-LL in OD (Figure 1f). The only correlation that was significant across both LOAD and OD was between PLA 2 activity and NAPE and only in the NP fractions. PLA 2 activity in LOAD negatively correlated with NAPE in NP (ρ = −0.47, p < 0.04) (Figure 1g). NP NAPE levels negatively correlated (ρ = −0.74, p < 0.05) with the activity of PLA 2 (Figure 1h).
We also determined the correlation of SP levels in CSF fractions from OD participants with aSMase or nSMase activities. Aβ 42 /Tau ratio was significantly related to nSMase activity (ρ = 0.68, p < 0.04). NP-SM was negatively related to the activity of aSMase (ρ = −0.68, p < 0.05), and NP dhCM was positively correlated with the activity of nSMase (ρ = 0.70, p < 0.03).
We next determined whether enhanced lipase activity accounted for metabolic differences between LOAD and OD. PLA2 activity correlated with several SF and NP lipids in CSF from OD participants. For OD samples, LPC in SF positively correlated with PLA2 activity in CSF (ρ = 0.81, p < 0.03), but no such correlation was evident for LOAD. Similarly, xPE levels in SF correlated with PLA2 activity (ρ = 0.86, p < 0.02) in OD but not in LOAD. PLA2 activity in LOAD did not correlate with NP PC (Figure 1a), while PLA2 activity positively correlated with NP PC in OD (Figure 1b).

Discussion
Our objective is to determine if there are any differences in lipid metabolic pathways in CSF from subjects with LOAD compared with those with OD. In this study, the OD population is made up of participants with DLB, FTD, VaD, and mixed dementia that can often be clinically differentiated from LOAD by the earlier age of onset, step-wise progression, and shorter survival time [48]. OD subjects have a median survival time from symptom onset of 6 ± 1.1 years compared with LOAD subjects who have a survival time from presentation of 11.7 years [56]. In our age-matched study population, duration of dementia is shorter in OD than in LOAD (Table 1). Given the accelerated rate of progression and death in OD, the degree of lipid metabolic abnormality may be different between OD and LOAD. Our study confirms this by demonstrating changes in GP, SP, and enzymes that metabolize important inflammatory and apoptotic lipids.
Examining GP classes, the only significant variation was PS that almost tripled in the SF of OD compared with LOAD. Typically found in the inner leaflet of plasma membranes and mitochondrial-associated membranes, PS only constitutes 2-10% of total lipids [57,58]. Although present in small amounts, PS is crucial in the signaling apoptosis [59]; cells externalize PS to the outer leaflet as a recognition signal to phagocytic cells or to initiate the blood clotting cascade [60]. The increase in PS in SF may indicate accelerated neurodegeneration in OD compared with LOAD study participants: OD brain cells exposed to higher PS levels are undergoing apoptosis faster than LOAD. Higher PS may account for the earlier clinical manifestation and faster cognitive decline in participants with OD.
Our data also indicate that there is a disruption in GP metabolism in LOAD and OD that may impact Aβ 42 and Tau metabolism. In regards to the correlation in LOAD between Aβ 42 and LPC/PC in SF, alteration in lipid metabolism may influence the proper clearance of Aβ 42 , resulting in the pathological features associated with Aβ 42 aggregation [61]. The indication, then, is that those with higher levels of Aβ 42 in CSF have a lower disturbance of lipid metabolism. In contrast, highly perturbed lipid membranes do not only result in abnormal clearance of Aβ 42 but increase the susceptibility of amyloid precursor protein (APP) to digestion by secretases [62,63]. We also found that higher levels of Tau, that indicate pathology found in dementia, positively correlated with LPC in both LOAD and OD CSF samples. Since LPC is derived from PC by PLA 2 -catalyzed lipolysis, induces myelination, and disrupts the blood-brain barrier [64], these data suggest that disruption of membrane structure is associated with dementia. Our data are in agreement with several studies showing that many PLA 2 isoforms contribute to AD pathology [23,[65][66][67].
SPs are another group of lipids essential in brain function and implicated in AD pathology [53,68]. Of the major SP classes detected in CSF, only the levels of CM was higher in NP of OD compared with LOAD. As a critical component for the initiation of apoptosis and autophagy, CM acts on several cells, including neurons [31,69,70]. Thus, higher levels of CM can cause increased rates of apoptosis in neural cells in OD compared with LOAD. This faster cell death would cause the clinical presentation to be identified earlier in OD compared to LOAD participants.
Further evidence for a role of SPs in OD is shown when Tau levels negatively correlate with CM and SM in SF. How Tau interacts with lipids has not been widely studied, but one can conclude that SP metabolism associates with Tau and this interaction is different in LOAD compared with OD.
Several enzymes modify GPs and SPs to form lipid mediators of inflammation or lipids that modify neuronal function. PLA 2 is notable for its role in Alzheimer's disease and other neurological disorders, due to the release of the inflammatory precursor arachidonic acid [67,71]. Once released, cyclooxygenase, lipoxygenase, and cytochrome p450 convert arachidonic acid to bioactive lipid mediators of inflammation. These PLA 2 -and COX-derived products are crucial in regulating an inflammatory response in LOAD [72]. Inflammatory cytokines can induce the synthesis of PLA 2 and also stimulate the de novo synthesis of ceramide and PAF [73][74][75][76]. In our study, PLA 2 activity in CSF from OD participants correlated with levels of several GPs; LPC and xPE in SF, and PC, PAF, NAPE, and LPC/PC in the NP fraction. PAF is an inflammatory GP produced and metabolized through two variations of PLA 2 -cytosolic and Lp-PLA 2 [77]. LPC interacts with inflammatory cells to increase cytokine production, such as tumor necrosis factor alpha (TNF-α) and interleukin (IL)-6 [78]. LPC appears to be a direct reflection on the intensity of reactive oxygen species present at the inflammatory site and can induce localized demyelination in rats [77,79]. A higher ratio of LPC/PC indicates that greater inflammatory markers are present in the CSF of OD patients. Correlation of PLA 2 with these lipids suggests that hydrolysis may be linked with destructive neuro-inflammation in subjects with OD.
NAPE is a significant component of intracellular membranes and has been shown to accumulate under pathological conditions that involve the degradation of membranes [80,81]. NAPE maintains proper physiological membrane configurations [81]. As NAPE accumulates in the brain tissue of LOAD and OD, fewer NAPE molecules will be removed in the CSF. The negative correlation between NAPE and PLA 2 observed in LOAD and OD reflects a disturbance in neuronal membrane lipids.
The activities of aSMAse and aSMase/nSMase ratio were higher in CSF from subjects with OD compared to LOAD. aSMase and nSMase are both involved in the degradation of SM to form CM, a known pro-apoptotic biolipid [31,82,83]. Functional consequences of SM degradation include apoptosis, growth arrest, and decreased protein secretion [84]. OD also demonstrated a significant negative correlation between aSMase activity and NP SM. The higher aSMase activity in OD compared with LOAD may account for elevated CM levels. Thus, higher aSMase activity in OD can result in faster apoptosis than in LOAD. A second positive correlation is noted between nSMase and dhCM. dhCM is a precursor to CM, and increased levels of dhCM are associated with oxidative stress [34,85]. Therefore, oxidative stress linked with lipid metabolic dysregulation may associate with AD pathophysiology.

Conclusions
This pilot cross-sectional study shows that lipid metabolism in CSF fractions differs between LOAD and OD. The small OD sample size limits our ability to examine previously described lipidomic variations between mixed dementia phenotypes [86]. These studies of lipid metabolism can provide insight into the dysfunction of metabolic pathways that can be used for diagnosis or to target AD therapy [87,88]. Replication with bigger sample size and consideration of different age of onset for defined dementia variants will be needed to associate known clinical phenotypes with specific lipid metabolic changes in CSF fractions. Acknowledgments: Study participants generously donated their time and CSF samples. Fuller Seminary Neuropsychology graduate students assisted in the diagnosis of AD. Sherri Lee provided clerical help and managed the study participants.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.